Aldert Vrij's research program starts from a premise that inverts the usual approach. If liars cannot be reliably detected by observation, stop observing and start intervening: make lying harder, and the difference between liars and truth-tellers will grow large enough to see.
The premise that lying is harder is well founded. A liar must construct a plausible account rather than retrieve one, monitor the interviewer for signs of belief, suppress the true version which keeps intruding, remember what they have already said in order to stay consistent, and control their own behavior because they know they are being watched. A truth-teller does one thing: remembers.
So the technique is to add load and see who breaks.
The 2008 study is the cleanest demonstration. Participants who had either committed a mock crime or not were interviewed, some in the ordinary way and some asked to recall the event in reverse order — starting at the end and working backwards. The interviews were then shown to observers who judged truthfulness.
Observers watching the standard interviews performed at the usual near-chance level. Observers watching the reverse-order interviews performed significantly better, with accuracy rising into the sixties. The liars in the reverse-order condition showed markedly more cues to deception than liars in the standard condition — the manipulation had made the difference visible rather than making the observers better.
Reverse order works because of what supports each kind of account. A fabricated narrative is built on a script — a chronological structure that generates plausible content in order. Running it backwards destroys the scaffolding, and the account has to be reconstructed on the spot. A genuine episodic memory is not stored as a script; it is a network of associated details that can be entered at any point, so traversing it non-linearly is effortful but possible.
Other load manipulations in the same program: asking the interviewee to maintain eye contact throughout, which is demanding and interferes with fabrication; asking a second task to be performed simultaneously; and the unanticipated questions of the next chapter.
Two important cautions.
The technique is not a lie detector and its effects are moderate. Accuracy in the sixties is a real improvement over fifty-four percent and is not certainty.
And it works only if the load is differential. A manipulation that is equally hard for everybody adds noise rather than signal — and there are populations for whom the baseline load is already high: people being interviewed in a second language, people with cognitive impairments, the very tired, the traumatized. Applying a load manipulation to them produces degraded performance that has nothing to do with deception, which is chapter 97's failure mode arriving through a more sophisticated door.
Cross-ref: unanticipated questions chapter